Studies suggest that endotoxin levels in offices and homes may be associated with sick building syndrome or increased asthma severity. However, reported endotoxin levels in these studies were highly inconsistent, suggesting incompatible measurements from various laboratories. Therefore, an investigation of Limulus assay for endotoxin in house dust was undertaken. Interference with the assay was common and could produce endotoxin estimates varying by a factor of > 100, depending on the dilution used. Analysis of dose-response curves allowed detection of two types of interference: dilution-dependent and dilution-independent. Dilution-dependent interference persisted when samples were reassayed, but valid estimates could be obtained by appropriate dilution and data analysis. Valid estimates could not be obtained from assay showing dilution-independent interference. However, dilution-independent interference was frequently overcome by repeating the assay. Estimates based on a single sample dilution produced reasonable results on average when a dilution factor of 900 was used. However, the results were less precise from a single dilution (median coefficient of variation [CV] 40%) than from the complete dose-response curve (median CV 23%) and ranged from 11 to 240% of the estimates based on complete dose-response data. A comparison of extraction methods found that sonication in a phosphate-triethylamine buffer produced greater endotoxin activity than extraction in buffer with addition of saponin or sodium dodecyl sulfate. Endotoxin activity was stable for 8 to 10 weeks in dust samples stored at 4 and -20 degrees C but was not stable in extracts. The lot of Limulus amebocyte lysate reagent and method of detecting a response in the kinetic Limulus assay had significant effects on endotoxin estimates.